There is a growing concern that use of thiazolidinediones (TZDs) is associated with congestive heart failure (CHF) (1), though the causality of this relationship has not been established. Commentaries and case reports about this potential side effect have frequently appeared in the press, but peer-reviewed, large-scale epidemiologic studies have been absent. A few European nations have already limited the use of TZDs despite the absence of empirical evidence. A recently published, longitudinal observational study by Delea et al. (2) reported that initiation of TZDs was associated with an increased risk of incident CHF (hazard ratio 1.7; P < 0.001). While we welcome this first population-based study of TZD use and CHF, we have serious concerns regarding the methodology employed.
Our primary concern with the study by Delea et al. (2) is the potential for residual confounding by indication or severity (3). For example, this study did not measure or adjust for levels of glycemia, a known risk factor for CHF in diabetes (4). In a time-to-event analysis, the authors compared patients who newly initiated TZDs with those who did not newly initiate TZD therapy while adjusting for other maintained diabetes therapies. Although patients may switch therapies due to side effects, most patients initiate intensified therapy because they have failed to maintain adequate glycemic control with previous regimens (5). Thus patients who initiate a new diabetes therapy, particularly a therapy not considered to be first line, would likely have poorer glycemic control and more advanced diabetes than those who maintain previous therapy and therefore may be at greater risk for CHF.
We are currently conducting a long-term study, funded by the American Diabetes Association, of TZD use and CHF in the Kaiser Permanente Northern California Diabetes Registry. In preliminary analyses, we found a substantially elevated prevalence of several markers of disease severity, including poor glycemic control (HbA1c >9.5%), among patients initiating new diabetes medications relative to those not starting new therapies. Moreover, compared with those initiating other therapies, TZD initiators had a higher prevalence of many CHF risk factors, including the greatest prevalence of ischemic heart disease, hypertension, elevated urinary albumin excretion, elevated serum creatinine, microalbuminuria, and obesity; the poorest glycemic control; and the lowest mean HDL cholesterol levels. TZD initiators were those most likely to also be prescribed medications for dyslipidemia and hypertension and had the greatest outpatient and inpatient utilization. Thus, TZDs were initiated more frequently in diabetic patients with more advanced disease. Since there currently are no generic TZDs, these expensive therapies are likely being reserved for more severe or advanced cases of diabetes.
Delea et al. (2) compared CHF incidence in TZD initiators with that among all other patients, a comparison that is even more biased than the comparison of TZD initiators to initiators of other diabetes therapies. Their comparison group is primarily comprised of those maintaining rather than initiating therapies and is thus healthier. These authors acknowledge that TZD initiators were sicker than subjects in their comparison group. Substantial imbalance in disease severity between exposure groups makes observational studies particularly vulnerable to bias, especially in the absence of thorough statistical adjustment. The authors relied on adjustment for prevalent conditions reported in the previous 12 months, including concurrent medication use and processes of care. While this may seem to be adequate at first glance, the impact of residual confounding remains unclear since important clinical adjusters, such as levels of glycemic control and markers of disease severity, measured over a longer time frame were not available. Additional uncertainties remain. If TZDs confer increased risk for CHF, one might expect a dose-response effect. No such evidence was reported in the study of Delea et al.
Even the largest well-designed premarketing trials often fail to uncover serious side effects caused by new therapies (6). Spontaneous adverse drug reaction reports from postmarketing surveillance are subject to overinterpretation given the often atypical clinical characteristics of cases (confounding), unawareness of the population background rate, and exaggerating effect of media focus, underscoring the importance of large epidemiologic studies to estimate the risk of adverse events associated with drug use (7–9). Although it is well accepted that TZDs may cause volume expansion and peripheral edema (10), the association between long-term TZD utilization and increased CHF risk requires further evaluation (11). Delea et al. should be commended for taking the first step in examining the potential for TZD side effects. It is clear that these authors have conducted the best possible analysis with the available data and fully acknowledge the potential limitations of their study design. However, we cannot exclude residual confounding as an explanation for these authors’ findings; any newly initiated diabetes therapy might be associated with elevated CHF risk. We therefore caution that the findings of Delea et al. should not be interpreted as causal. Changes in clinical recommendations for TZDs should be based on solid evidence; we do not think this suffices.